Watermarking and data hiding have been extensively researched (e.g., IEEE ICIP, ISCAS, ICMCS, ICASSP, and SPIE special sessions and conferences for the past two decades). Due to the significant broadband consumer electronics applications, such watermarking provides an underlying enabling technology for digital rights management, steganography, watermarking, copy protection, copyright protection, traitor tracing, and/or IP protection.
Conventional watermarking extraction techniques include (i) taking a large number (e.g., 1000) of the highest amplitude discrete cosine transform (DCT) coefficients in an image (or video frame), (ii) averaging 8×8 blocks of an image (essentially the equivalent to taking the DC 8×8 transform coefficients), (iii) subtracting the original copy and projecting the remaining copy onto a subspace, and (iv) finding salient points and Delaunay triangulating the salient points for representation as a graph.
Among conventional watermarking extraction techniques, working directly on a compressed video stream is effective computationally. Inserting a robust watermark into a compressed video stream involves subtle manipulation of standard high bitrate syntax elements of a stream. Information may be directly inserted into a video in the pixel domain prior to compression. However, by inserting video in the pixel domain prior to compression, a much higher data processing rate (typically 50 to 100 times higher) is needed. A compression process that follows inserting video in the pixel domain is inefficient since the information quality weakens. It is computationally efficient to insert the information into the compressed stream after all easily accessible sources of redundancy have been removed from the data.
Typically, when the bitrate is high and the syntax element that is in use includes less perceptually visible distortion, more information can be inserted into the compressed stream. This approach has led to the preference shown in the past towards inserting information into transform coefficients, particularly the low-frequency (the lowest frequency of which is the DC coefficients).
Conventional approaches including using transform coefficients that are quite complex and need processing of the majority of an entire compressed bitstream to effectively insert information. In addition, with conventional approaches, robustness, security, and imperceptibility could be improved.
Watermark/fingerprint insertion and extraction processes are computationally demanding. Watermarks and fingerprints inserted for traitor tracing should be robust, secure, and imperceptible and should not be removable without distorting the video. Conventional approaches mainly differ from each other by the models used to control fidelity, robustness, bitrate, and error rates.
It would be desirable to provide a method and/or apparatus for video watermarking and steganography that improves robustness, security, and imperceptibility of the watermarking insertion and extraction process.